yield checksum every N duration

[morgo]

To start a checksum, spirit acquires a metadata lock (FLUSH TABLES WITH READ LOCK) and then starts repeatable read transactions for the N threads that are required to checksum. This can lead to long HLL values if the checksum process takes a long time.

Historically, the checksum was non-resumable, but this has long been fixed. Checksums are checkpointed, and so it's possible to resume progress.

This is a feature request to basically yield the checksum every N duration (lets start with a default of 24 hours). The downside of yielding the checksum is it requires the metadata lock to be acquired again, which is not ideal for some systems. In practice however, we will likely want to lower that default.

[morgo]

@Kiran01bm do I recall correctly that you expressed interest in working on this?